{"date_published":"2019-12-20T00:00:00Z","status":"public","conference":{"end_date":"2020-05-01","location":"Virtual","name":"ICLR: International Conference on Learning Representations","start_date":"2020-04-26"},"quality_controlled":"1","extern":"1","_id":"14184","scopus_import":"1","citation":{"mla":"Locatello, Francesco, et al. “Disentangling Factors of Variation Using Few Labels.” <i>8th International Conference on Learning Representations</i>, 2019.","ista":"Locatello F, Tschannen M, Bauer S, Rätsch G, Schölkopf B, Bachem O. 2019. Disentangling factors of variation using few labels. 8th International Conference on Learning Representations. ICLR: International Conference on Learning Representations.","ieee":"F. Locatello, M. Tschannen, S. Bauer, G. Rätsch, B. Schölkopf, and O. Bachem, “Disentangling factors of variation using few labels,” in <i>8th International Conference on Learning Representations</i>, Virtual, 2019.","short":"F. Locatello, M. Tschannen, S. Bauer, G. Rätsch, B. Schölkopf, O. Bachem, in:, 8th International Conference on Learning Representations, 2019.","apa":"Locatello, F., Tschannen, M., Bauer, S., Rätsch, G., Schölkopf, B., &#38; Bachem, O. (2019). Disentangling factors of variation using few labels. In <i>8th International Conference on Learning Representations</i>. Virtual.","ama":"Locatello F, Tschannen M, Bauer S, Rätsch G, Schölkopf B, Bachem O. Disentangling factors of variation using few labels. In: <i>8th International Conference on Learning Representations</i>. ; 2019.","chicago":"Locatello, Francesco, Michael Tschannen, Stefan Bauer, Gunnar Rätsch, Bernhard Schölkopf, and Olivier Bachem. “Disentangling Factors of Variation Using Few Labels.” In <i>8th International Conference on Learning Representations</i>, 2019."},"title":"Disentangling factors of variation using few labels","article_processing_charge":"No","oa_version":"Preprint","department":[{"_id":"FrLo"}],"type":"conference","date_updated":"2023-09-12T07:01:34Z","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","author":[{"id":"26cfd52f-2483-11ee-8040-88983bcc06d4","full_name":"Locatello, Francesco","last_name":"Locatello","orcid":"0000-0002-4850-0683","first_name":"Francesco"},{"full_name":"Tschannen, Michael","first_name":"Michael","last_name":"Tschannen"},{"last_name":"Bauer","first_name":"Stefan","full_name":"Bauer, Stefan"},{"last_name":"Rätsch","first_name":"Gunnar","full_name":"Rätsch, Gunnar"},{"last_name":"Schölkopf","first_name":"Bernhard","full_name":"Schölkopf, Bernhard"},{"full_name":"Bachem, Olivier","last_name":"Bachem","first_name":"Olivier"}],"day":"20","language":[{"iso":"eng"}],"publication":"8th International Conference on Learning Representations","main_file_link":[{"open_access":"1","url":"https://arxiv.org/abs/1905.01258"}],"month":"12","date_created":"2023-08-22T14:06:37Z","publication_status":"published","oa":1,"abstract":[{"lang":"eng","text":"Learning disentangled representations is considered a cornerstone problem in\r\nrepresentation learning. Recently, Locatello et al. (2019) demonstrated that\r\nunsupervised disentanglement learning without inductive biases is theoretically\r\nimpossible and that existing inductive biases and unsupervised methods do not\r\nallow to consistently learn disentangled representations. However, in many\r\npractical settings, one might have access to a limited amount of supervision,\r\nfor example through manual labeling of (some) factors of variation in a few\r\ntraining examples. In this paper, we investigate the impact of such supervision\r\non state-of-the-art disentanglement methods and perform a large scale study,\r\ntraining over 52000 models under well-defined and reproducible experimental\r\nconditions. We observe that a small number of labeled examples (0.01--0.5\\% of\r\nthe data set), with potentially imprecise and incomplete labels, is sufficient\r\nto perform model selection on state-of-the-art unsupervised models. Further, we\r\ninvestigate the benefit of incorporating supervision into the training process.\r\nOverall, we empirically validate that with little and imprecise supervision it\r\nis possible to reliably learn disentangled representations."}],"year":"2019","external_id":{"arxiv":["1905.01258"]}}